Detection of Soccer Goal Shots Using Joint Multimedia Features and Classification Rules

Paper Abstract:

As digital video data becomes more and more pervasive, the issue of mining information from video data becomes increasingly important. In this paper, we present an effective data mining framework for automatic extraction of goal events in soccer videos. The extracted goal events can be used for high-level indexing and selective browsing of soccer videos. The proposed multimedia data mining framework first analyzes the soccer videos by using joint multimedia features (visual and audio features). Then the data pre-filtering step is performed on raw video features with aid of domain knowledge, and the pre-filtered data are used as the input data in the data mining process using classification rules. The proposed framework fully exploits the rich semantic information contained in visual and audio features for soccer video data, and incorporates the data mining process for effective detection of soccer goal events. This framework has been tested using soccer videos with different styles as produced by different broadcasters. The results are promising and can provide a good basis for analyzing the high-level structure of video content.

Fig. 1. The architecture of the proposed framework

Fig. 2. The workflow of the proposed shot change detection method

 
(a)                       (b)

(c)                       (d)

Fig. 3. (a) a sample frame from a goal shot (global view); (b) a sample frame from the cheering shot following the goal shot for (a); (c) object segmentation result for (a); (d) object segmentation result for (b).

Fig. 4. The histogram of the candidate grass values for a 20-minute long soccer video. Two peaks correspond to two major types of shooting scales in the video data – global and close-up.

Fig. 5. Detected grass areas (black areas) for 3 sample video frames from different types of shots.

(a)                 (b)               (c)

(d)                 (e)               (f)

Fig. 6. Goal shots followed by close shots: (a)-(c) three consecutive shots in a goal event. (b) is the close shot follows (a) the goal shot; (d)-(f) another goal event and its three consecutive shots, (f) is the close shot follows (d) the goal shot.

Fig. 7. Workflow of mining goal shots in soccer video

Office: ECS 362
Phone: 305-348-3480
Fax: 305-348-3549
E-mail: chens@cs.fiu.edu
 
Associate Professor
School of Computer Science, 1999-Present
 
Director
Distributed Multimedia Information System Laboratory
 
Associate Director
The Center for Advanced Distributed System Engineering
 
Florida International University,
The State University of Florida at Miami